News | Nvidia supplier nabs Texas site for $1.3 billion manufacturing hub - CoStar

This $1.3B Texas facility is a structural admission: AI compute's bottleneck has migrated off-die. The constraint is no longer GPU silicon but the packaging and interconnect layer determining real-world throughput. Technical ripple: With 3nm yields plateauing, the performance ceiling has shifted to 2.5D/3D advanced packaging and high-frequency substrate. This facility almost certainly targets CoWoS-class assembly or HBM interconnect. Once ramped, it accelerates Nvidia's GB200/GB300 cadence while tightening the window for AMD's MI400 and Intel's Gaudi successors. Compliance calculus: Choosing Texas over Arizona signals optimization for energy cost and state tax incentives over federal CHIPS subsidies. The hidden risk is upstream—if critical substrates still route through Taiwan, China or Southeast Asian nodes, any export-control escalation could strand a partially built line. At $1.3B sunk, a supply rupture means an 18-month rebuild. Competitive response: Expect AMD and Intel to secure secondary capacity or accelerate in-house packaging within two quarters. If TSMC perceives its CoWoS output monopolized by one customer, it will accelerate US-based advanced packaging—reshaping where AI chips are actually finished. Outlook: Within 18 months, "packaging is compute" becomes the operating assumption. Nvidia's moat extends from CUDA to a vertical silicon-plus-package-plus-interconnect loop. The $1.3B isn't buying a factory—it's buying a time window competitors cannot replicate.